Physx

[Vale]

Mate has a Sapphire Radeon 4850, can this card do Physx like a 9800GTX or not. If not is it worth him getting a Physx card to do the job. On my gtx it has a Physx tab and settings in the drivers, his does not so I assumed ATI never did Physx?

 

[Ben M]

No it can't do physx calculations. No it isn't worth getting another card to do physx, because it wouldn't work (it works in Windows 7, but you have to have a full on fight with drivers)

 

[reflux]

No, and no. Physx is a total waste of time right now anyway -

Until we get a standard for GPU accelerated physics that all GPU vendors can use or until NVIDIA can somehow convince every major game developer to include compelling features that will only be accelerated on NVIDIA hardware, hardware PhysX will be nothing more than fancy lettering on a cake.

http://anandtech.com/video/showdoc.aspx?i=3539&p=12

 

[Vale]

Ok, thanks guy's.

 

[LoadsaMoney]

As above, but it can be paired with a Nvidia card that can do the PhsyX, only works in XP or Win 7, but also as said above, its not worth it yet.

 

[Rroff]

Yeah as loadsmoney said you can get a cheapy (tho I'd advise 8800GS, 8800GT or better, 9600GT with only 64SPs isn't the best card tho it does ok) nvidia card to do physx on XP and win7...

However unless you play one of the few games that make much use of physx theres little point at the moment.

 

[manic_man]

It's a brilliant software library that is used far more extensively than people realise.. however, its ability to hardware accelerate on modern NVIDIA GPUs is limited, so on the whole most of the time the library itself is being used, having a card capable of it wont make much difference.

It's great from a programmers perspective though (not that will be of much interest to the average Joe )

 

[Rroff]

It's a brilliant software library that is used far more extensively than people realise.. however, its ability to hardware accelerate on modern NVIDIA GPUs is limited, so on the whole most of the time the library itself is being used, having a card capable of it wont make much difference.

It's great from a programmers perspective though (not that will be of much interest to the average Joe )

This is the thing the average joe hasn't seen what its capable of, so writes it off, developers know what its capable of and would like to get it into games... but everyone wants the proof before the pudding... so its a vicious circle.

The current generation hardware physx is quite capable and fully functional, other than its still a little bit low on the active bodies count for my tastes.

Unfortunatly no developer is going to spend time implementing features that aren't incidental ones and can be turned off without gameplay consequences, and the features that would make an impression on people would need to be much more integrated with the game so with the current state and attitudes of the market... everyone loses out.

 

[manic_man]

This is the thing the average joe hasn't seen what its capable of, so writes it off, developers know what its capable of and would like to get it into games... but everyone wants the proof before the pudding... so its a vicious circle.

The current generation hardware physx is quite capable and fully functional, other than its still a little bit low on the active bodies count for my tastes.

Unfortunatly no developer is going to spend time implementing features that aren't incidental ones and can be turned off without gameplay consequences, and the features that would make an impression on people would need to be much more integrated with the game so with the current state and attitudes of the market... everyone loses out.

The major missing article from the current incarnation is that of hardware accelerated rigid bodies, supposedly to come in the next iteration of the SDK, so fingers crossed. And yeah you're right, the 4k shape limit per scene when using a HW scene manager is quite limiting (to the point that I have devoted quite a bit of my PhD to working around this!!) My other gripe is that if you run on a PPU, they wont gurantee determinism due to the vector based approach of the old PPU, similarly the same can be said of the inbuilt multithreading model.

Certainly its solver accuracy and general speed and functionality far surpasses that of its rivals.

I wonder how many games are out there using PhysX purely in software mode as its contact detection algorithms and raycasting methods are far better than a small team of devs could likely put together in a few months.

I'm currently trying to drag the real-time physics engine ethos into the world of scientific modelling and am using PhysX as my examplar case, its not ideally suited, but certainly makes a good argument when presented correctly.

I'm rarely impressed with software libraries, they are what they are, but I find PhysX a very neat piece of work indeed.

Sorry taken this thread a bit off track! I think the other responses here so far have been correct though, for now, dont worry about getting a dedicated processing card. IMO you may even see a PhysX SDK compatible with not only CUDA but also OpenCL in the future as well, so ATIs cards will be able to get in on the game natively. Could be wrong though oc

 

[Rroff]

ouch I thought RBs were hardware accelerated, that would seem like most of the point of hw physics...

I spend most of my time when I do anything with physics working with tokamak (ok for simple stuff but none too stable) and havok (which I hate but is quite well rounded).

 

[manic_man]

They are HW accelerated on the old PPUs, but at the moment, all RB stuff is done on software.

Supposedly they have had to effectively re-write their RB code to work with CUDA, what they are producing is effectively based around the "particles" example that can be found in the CUDA SDK. The release of the SDK with the HW RB stuff was due a few months ago, so it should be out anytime soon.

What I do like about PhysX is the segmentation that can occur. With the ability to have scenes and compartments and the like. Its a really nice hyriechical system that makes lots of sense from the game producers point of view, i'm making use of the idea of siulation segmentation in order to get around the limitations of real-time physics engines (i.e. many simulations all working in parallel together)

 

[KraniX]

http://www.firingsquad.com/hardware/mirrors_edge_physx_performance/default.asp

Interesting article using Mirror's Edge to compare some cards and the difference between GPU, CPU and a PPU options.